Our Review Process
Our review process comes down to a few core steps.
1. Data Ingestion
We collect both manufacturer and user-generated review data. This involves crawling the internet to gather product specifications, installation manuals, tutorials, and customer reviews.
Our process combines automated systems with human-led verification to ensure data quality and reliability. This step is critical—our conclusions can only be as accurate as the information they are built upon.
The types of data we collect include:
- Product specifications (dimensions, features, functions)
- User reviews & testimonials (real-world experiences)
- Product imagery (official photos, user-uploaded images)
- Videos & demonstrations (installations, usage walkthroughs)
- Technical diagrams & manuals (official PDFs, exploded views)
- Comparison charts (feature-by-feature breakdowns)
- Ratings & scores (aggregated from multiple platforms)
2. Data Cleaning
One of the biggest challenges in our process is the inconsistency and inaccuracy of source data. Sometimes, even manufacturers publish conflicting specifications. In other cases, fake or misleading reviews distort the picture.
We address this by continually refining and reapplying our cleaning methods—not only to new data but also to existing datasets, ensuring our information stays as accurate and up-to-date as possible.
3. Data Enrichment
Once we have a clean, verified dataset for a product, we begin enriching it. This means structuring the data into a consistent format and applying feature-based rankings.
Our enrichment process is built on a human-designed system that interprets, scores, and prioritizes product features in a way that reflects real-world use and consumer priorities. We measure not only the sentiment expressed in user tested data sets, but also account for the materials, manufacturing processes, and the consistency of each brand's quality. Together, these factors form the foundation of the scores you see in our product reviews.
4. Score Assignment
Utilizing a scoring matrix fine-tuned by our in-house expert, we assign scores to:
- Design and materials
- Feature count
- Individual feature performance
Scores are derived from ranking objective materials, manufacturing process, design, and by extracting qualities of features from re-occurring user-tested review data.
The Role of AI
Yes, we do use AI—but not in the way most people assume. We don't simply hand our process over to a public or private large language model (LLM). Instead, we use AI as a controlled processing aid.
Our datasets are collected and manually screened by us first. LLMs are then applied within strict guardrails, limited to only the data we provide. This ensures that all results are grounded exclusively in the curated information we control—never in AI's "memory" or external data sources.